Here is the English translation of the provided paragraph while maintaining the original format: ---

Over the past two decades, language modeling technology has evolved from statistical models to neural models, with pre-trained language models (PLMs) demonstrating outstanding performance in NLP tasks. The emergence of large language models (LLMs) like ChatGPT marks the dawn of the era of artificial general intelligence, capable of automatically generating textual content and being applied across various domains. However, challenges remain in terms of comprehension, logical consistency, and reliability. OpenAI’s GPT-4 model has shown improvements in these areas. Technological advancements also bring ethical and societal concerns. As one of China’s largest social media platforms, comments on Sina Weibo reflect public attitudes toward trending topics. This study analyzes public discourse on ChatGPT and artificial intelligence by mining Weibo comments, employing the BERTopic model for thematic analysis and Baidu’s sentiment analysis tools, supplemented by visualizations to present public attitudes and focal points regarding AI. With technological progress, exploring public attention and sentiment toward LLMs holds significant importance.

1. Literature Review

This section reviews existing research on sentiment analysis and topic modeling, as well as studies related to ChatGPT. Sentiment analysis involves assessing emotional tendencies in text, with primary methods including lexicon-based approaches, machine learning, and deep learning—the latter achieving higher accuracy due to the integration of attention mechanisms. In topic modeling, the Latent Dirichlet Allocation (LDA) model is mainstream but struggles with short texts, whereas the BERTopic model has gained attention for its superior semantic interpretability. Research on ChatGPT can be categorized into three phases: technical principles, societal impact, and future trends, covering aspects such as technological innovation, social applications, and ethical concerns. Current studies on LLMs predominantly focus on technical aspects, with limited exploration of public sentiment. However, public acceptance and areas of interest in new technologies hold significant reference value for model development.

2. Research Methodology

This section introduces the topic-sentiment analysis method based on Weibo comment data. First, Python Scrapy was used to scrape Weibo comments related to ChatGPT and AI, followed by preprocessing steps such as word segmentation and stopword removal, resulting in 9,308 valid entries. Next, the BERTopic model was employed to investigate public focus areas, while Baidu’s sentiment analysis tool was applied to assess emotions within each topic. The BERTopic model leverages pre-trained BERT for topic modeling, offering stronger semantic interpretability. The topic analysis involved converting text into vector representations using sentence_transformers, UMAP, and K-means clustering, followed by keyword extraction via class-based TF-IDF and refinement using the Maximal Marginal Relevance algorithm. Baidu’s sentiment analysis tool, trained on deep learning, automatically classifies text into positive, negative, or neutral sentiments. Additionally, this study extracted frequencies of positive and negative words using a sentiment lexicon and generated word clouds to visualize public concerns regarding LLMs.

3. Results Analysis

This chapter reveals key public opinions and attitudes toward ChatGPT through word frequency analysis, topic modeling, and sentiment analysis. Word frequency analysis identified "artificial intelligence" and "model" as high-frequency terms, reflecting ChatGPT’s broad applications in technology. BERTopic analysis identified four major public concerns: ethics, education, technological development, and investment. In ethics, discussions centered on ChatGPT’s impact on moral judgment and its inherent vulnerabilities. In education, ChatGPT was seen as enhancing work and study efficiency but also raising concerns about academic misconduct. In technological development, public attention focused on ChatGPT’s applications and future prospects, alongside worries about job displacement. In investment, discussions revolved around ChatGPT’s predictive capabilities and investment potential. Sentiment analysis showed predominantly positive online sentiment toward ChatGPT, with words like "innovation" and "breakthrough" indicating optimism about AI’s future, while terms like "vulnerability" and "leakage" reflected concerns about misinformation and data security.

4. Conclusion and Outlook

This chapter summarizes the analysis of public attitudes toward LLMs based on Weibo comments, revealing overall optimism alongside concerns about security, unemployment, and content authenticity. Recommendations include enhancing professional skills, fostering critical thinking, and emphasizing ethical considerations to adapt to technological advancements, mitigate over-reliance, and ensure reliable information.

--- The translation preserves the original structure, headings, and paragraph divisions while ensuring clarity and accuracy in English. No additional descriptions or modifications outside the original text have been added.